Nuclear power plant fault diagnosis method and device

Abstract

The invention discloses a nuclear power plant fault diagnosis method and device, and the method comprises the steps: collecting nuclear power plant data, training and constructing an edge end model group and a cloud model group according to the nuclear power plant data, and the nuclear power plant data comprises operation regulations, real-time operation data and historical operation data; deploying the edge end model group to an edge computing node near equipment in a plant, and deploying the cloud end model group to a cloud platform; inputting the real-time operation data of the nuclear power plant into the edge end model group, detecting a fault type, reconstructing sensor parameters if the fault type is sensor abnormality, and uploading an abnormality report to the cloud platform if the fault type is system abnormality, and reasoning and tracing the abnormal report through the cloud model group, and pushing fault information and a decision scheme.

Description

Technical Field

[0001] The present invention relates to the field of nuclear power plant fault diagnosis, and particularly to a nuclear power plant fault diagnosis method and device. Background Art

[0002] With the development of advanced technologies, informatization, artificial intelligence, and intelligence are the future development trends of nuclear power systems. However, due to the complex nuclear power plant systems, numerous instruments, and scattered layouts, there are a large number of alarms when faults occur, and it is difficult for operators to identify the source of the faults in a short time. If an unplanned reactor shutdown is caused by failure to respond promptly to off-normal conditions, it will also cause huge economic losses to the nuclear power plant.

[0003] Current fault diagnosis systems often rely on centralized cloud computing or single-edge computing. However, relying solely on cloud computing will lead to data transmission delays and cannot meet the high requirements of nuclear power plants for real-time performance and reliability; while relying only on edge computing, limited by computing power and storage resources, it is difficult to complete complex fault analysis and training of deep learning models, and there are problems such as poor real-time performance, low data processing efficiency, and insufficient system reliability; the traditional centralized cloud computing mode may also lead to problems such as data islands, data transmission delays, and analysis lags, and cannot meet the requirements of nuclear power plants for high-speed response and real-time accuracy of fault diagnosis functions; while a single-edge computing mode is limited by the computing power of edge devices and is difficult to complete complex fault analysis.

[0004] In the patent document CN117435894A, an intelligent GIS device fault location and detection method based on cloud-edge collaboration is disclosed. Using the idea of cloud-edge collaboration, through the combination of a digital twin model and an abnormal change pattern recognition method, comprehensive device monitoring, fault prediction, and precise positioning are achieved; by simulating the interaction relationships between the various modules of the GIS device through the digital twin model, a cloud-based large interaction detection model constructed using a graph neural network algorithm can monitor and analyze the interaction data of the GIS device in real time, and improve the accuracy and recall rate of interaction anomaly detection, but it does not solve the problems that fault diagnosis only through a single-sided model in the nuclear power field cannot meet the requirements of nuclear power plants for real-time performance and reliability and cannot accurately analyze and trace complex faults.

[0005] In the patent document CN114779739B, a fault monitoring method for industrial processes under cloud-edge-end collaboration based on a probabilistic graphical model is disclosed, including constructing a three-layer architecture for plant-level distributed fault monitoring and traceability in the cloud-edge-end collaborative environment for complex industrial processes, realizing dimensionality reduction and decomposition of complex data on the device side, performing local process monitoring of subsystems on the edge side, and performing subsystem information fusion on the cloud side to complete global fusion diagnosis of the entire industrial process. It solves the problem that in the nuclear power field, fault diagnosis only through a unilateral model cannot meet the requirements of nuclear power plants for real-time performance and reliability, and cannot accurately analyze and trace complex faults.

[0006] In summary, neither of the above two existing patents solves the problem that in the nuclear power field, fault diagnosis only through a unilateral model cannot meet the requirements of nuclear power plants for real-time performance and reliability, and cannot accurately analyze and trace complex faults. Summary of the Invention

[0007] Based on the above technical problems, the present invention proposes a fault diagnosis method and device for nuclear power plants, which solves the problem that in the nuclear power field, fault diagnosis only through a unilateral model cannot meet the requirements of nuclear power plants for real-time performance and reliability, and cannot accurately analyze and trace complex faults.

[0008] To achieve the above object, the present invention proposes a fault diagnosis method for nuclear power plants.

[0009] A fault diagnosis method for nuclear power plants includes:

[0010] Collecting nuclear power plant data, training and constructing an edge-side model group and a cloud-side model group according to the nuclear power plant data, where the nuclear power plant data includes operation procedures, real-time operation data, and historical operation data;

[0011] Deploying the edge-side model group to edge computing nodes near the equipment in the plant building, and deploying the cloud-side model group to the cloud platform;

[0012] Inputting the real-time operation data of the nuclear power plant into the edge-side model group to detect the fault type. If the fault type is sensor anomaly, reconstruct the sensor parameters. If the fault type is system anomaly, upload the anomaly report to the cloud platform, and perform reasoning and traceability on the anomaly report through the cloud-side model group, and push the fault information and decision-making scheme.

[0013] Further, training and constructing an edge-side model group and a cloud-side model group according to the nuclear power plant data includes:

[0014] Training based on the real-time operation data and the historical operation data to obtain a principal component dimensionality reduction model, a latent variable regression model, and an independent feature extraction model, and obtaining the edge-side model group through model integration.

[0015] Further, training and constructing an edge - side model group and a cloud - side model group according to the nuclear power plant data further includes:

[0016] Obtaining a knowledge graph of the nuclear power plant according to the operation regulations, training based on the real - time operation data and the historical operation data to obtain a signed directed graph model and a deep - learning confidence calculation model, and obtaining the cloud - side model group through model integration.

[0017] Further, obtaining the edge - side model group through model integration includes:

[0018] Setting a dynamic voter in the edge - side model group, and the dynamic voter makes a decision based on the results of the principal component reduction model, the latent variable regression model, and the independent feature extraction model.

[0019] Further, the dynamic voter makes a decision based on the results of the principal component reduction model, the latent variable regression model, and the independent feature extraction model, including:

[0020] After inputting the real - time operation data of the nuclear power plant, normalizing the results of the principal component reduction model, the latent variable regression model, and the independent feature extraction model.

[0021] Further, the dynamic voter makes a decision based on the results of the principal component reduction model, the latent variable regression model, and the independent feature extraction model, and further includes:

[0022] Inputting the historical operation data into the edge - side model group to obtain the threshold TH of each model respectively, and setting 0.25TH - 0.6TH as the standard threshold difference.

[0023] Further, the dynamic voter makes a decision based on the results of the principal component reduction model, the latent variable regression model, and the independent feature extraction model, including:

[0024] Judging for the real - time operation data. If the absolute value of the difference between the threshold TH of a single sensor and the mean of the threshold TH of other sensors exceeds 3 times the standard threshold difference, it is determined that the sensor is abnormal; if the absolute value of the difference between the threshold TH of all sensors and the threshold TH calculated based on historical normal operation data exceeds 3 times the standard threshold difference, it is determined that the system is abnormal;

[0025] When the output results of most models are determined to be the same abnormal type, it is directly determined to be that abnormal type. When the output results of each model are different, select the abnormal type determined by the model with the highest weight.

[0026] Further, if the fault type is sensor anomaly, reconstruct the sensor parameters, including:

[0027] Based on the historical operation data, correct the sensor parameters.

[0028] Further, after correcting the sensor parameters, it also includes:

[0029] Re - input the corrected sensor parameters into the edge - side model group.

[0030] Further, reason and trace the anomaly report through the cloud - side model group, including:

[0031] Based on the real - time operation data and the anomaly report, calculate the confidence level through the deep - learning confidence - level calculation model;

[0032] Reason the anomaly report through the knowledge graph to obtain a fault result set, and based on the fault result set, reason out a solution set through the signed - digraph model;

[0033] Determine the fault from the fault result set through the deep - learning confidence - level calculation model.

[0034] Further, determine the fault from the fault result set through the deep - learning confidence - level calculation model, including:

[0035] Through the deep - learning confidence - level calculation model, calculate and weight the faults in the fault result set to obtain the confidence level.

[0036] Further, push the fault information and decision - making solutions, including:

[0037] Push the fault information with the highest confidence level in the fault result set and its corresponding decision - making solution in the solution set to the front - end, and perform fault location on the relevant process system flow chart.

[0038] To achieve the above object, the present invention also proposes a nuclear - power - plant fault diagnosis device.

[0039] A nuclear - power - plant fault diagnosis device, characterized by including:

[0040] An acquisition and construction module, used to acquire nuclear - power - plant data, train and construct an edge - side model group and a cloud - side model group according to the nuclear - power - plant data, where the nuclear - power - plant data includes operation regulations, real - time operation data, and historical operation data;

[0041] A module deployment module, used to deploy the edge - side model group to an edge - computing node near the equipment in the plant building, and deploy the cloud - side model group to a cloud platform;

[0042] The inference and decision-making module is used to input the real-time operation data of the nuclear power plant into the edge model group to detect the fault type. If the fault type is sensor anomaly, the sensor parameters are reconstructed. If the fault type is system anomaly, the anomaly report is uploaded to the cloud platform, and the cloud model group is used to infer and trace the anomaly report, and push the fault information and decision-making scheme.

[0043] Based on the above technical solutions, the present invention has at least the following beneficial effects:

[0044] 1. The present invention provides a fault diagnosis method and device for a nuclear power plant. Through the collaborative architecture of the cloud and the edge, the advantages of the edge and the cloud are fully utilized, realizing the high efficiency and accuracy of nuclear power plant fault diagnosis; the edge model group is responsible for real-time data collection and preliminary processing, capable of quickly responding to abnormal situations, reducing data transmission latency, and ensuring the real-time nature of fault monitoring; the cloud model group, relying on its powerful computing and storage capabilities, conducts in-depth analysis and model training on complex faults, providing high-precision diagnostic results; the present invention also integrates an expert knowledge inference model and various data-driven algorithms, capable of analyzing and verifying faults from different perspectives, quickly locating possible fault types and fault sources; the data-driven model can effectively monitor abnormal data, distinguish sensor faults from system faults, and further evaluate the confidence level of the fault source, ultimately achieving accurate fault tracing; this multi-model fusion method not only improves the accuracy of fault diagnosis, but also can analyze the fault propagation path and predict the fault evolution trend, providing a scientific basis for taking preventive measures in advance.

[0045] 2. The present invention provides a fault diagnosis method and device for a nuclear power plant. By optimizing the data processing process and resource allocation through various data-driven algorithm models (such as PCA, PLS, ICA), the reliability and scalability of the system are significantly improved; on the one hand, after preprocessing the collected data, the edge model group only uploads the key data to the cloud, effectively reducing the data transmission volume, reducing the network bandwidth pressure and data transmission latency, thereby improving the overall reliability of the system; on the other hand, the collaborative architecture of the cloud model group enables the system to flexibly adjust the resource configuration according to the scale and data volume of the nuclear power plant, further calculating the fault confidence level through deep learning models (such as LSTM), and the edge sensor devices can be flexibly arranged according to actual needs, while the cloud resources can be dynamically expanded according to the computing and storage requirements, adapting to the application scenarios of nuclear power plants of different scales, showing good scalability.

[0046] 3. The present invention provides a nuclear power plant fault diagnosis method and device. By constructing an auxiliary decision-making knowledge graph and combining the operation data and operation procedures of the nuclear power plant, corresponding processing strategies and procedures can be quickly generated based on the fault diagnosis results, providing intuitive and scientific decision-making references for operators; the system uses a cloud model group for in-depth analysis and pushes decision-making information to the main control room and on-site maintenance personnel in real time to ensure information synchronization and improve decision-making efficiency and response speed; in addition, the system also has functions of fault traceability and fault propagation path analysis, which can help operators anticipate the fault evolution trend in advance, so as to take more reasonable countermeasures. The combination of such real-time monitoring, early warning and auxiliary decision-making functions can significantly reduce the probability of unplanned reactor shutdown in the nuclear power plant, reduce economic losses, and ensure the safe and stable operation of the nuclear power plant at the same time.

[0047] 4. The present invention provides a nuclear power plant fault diagnosis method and device. By clearly defining the process division of labor between the edge model group and the cloud model group, the problem of resource waste in the traditional centralized or single-edge computing mode is avoided; the edge model group focuses on real-time data processing, and the cloud model group is responsible for complex analysis and model training. This division of labor makes resource utilization more efficient; at the same time, the system supports flexibly selecting to deploy the data-driven model of the edge model group to the edge or the cloud according to the needs of the monitoring object, further improving the flexibility and adaptability of the system. This flexible deployment strategy can not only meet the needs of different application scenarios, but also dynamically adjust resource allocation according to the actual operation situation to ensure that the system can operate efficiently under different working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The accompanying drawings forming a part of the present invention are used to provide a further understanding of the present invention. The schematic embodiments and descriptions thereof of the present invention are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0049] Figure 1 The flowchart of a nuclear power plant fault diagnosis method according to an embodiment is shown;

[0050] Figure 2 The training flowchart of the principal component reduction model according to an embodiment is shown;

[0051] Figure 3 The training flowchart of the latent variable regression model according to an embodiment is shown;

[0052] Figure 4 The training flowchart of the independent feature extraction model according to an embodiment is shown;

[0053] Figure 5 The structural schematic diagram of a nuclear power plant fault diagnosis device according to an embodiment is shown;

[0054] Figure 6 The structural schematic diagram of a nuclear power plant fault diagnosis product according to an embodiment is shown;

[0055] Figure 7 The structural schematic diagram of an electronic device according to an embodiment is shown. Specific embodiments

[0056] It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments may be combined with each other. The present invention will be described in detail below with reference to the drawings and in combination with the embodiments.

[0057] The following further describes the present invention in detail with specific embodiments, and these embodiments should not be construed as limiting the scope claimed by the present invention.

[0058] Embodiment

[0059] To solve the problems that in the nuclear power field, fault diagnosis only through a unilateral model cannot meet the requirements of nuclear power plants for real-time performance and reliability, and cannot accurately analyze and trace complex faults, the present invention proposes a nuclear power plant fault diagnosis method and device.

[0060] To achieve the above object, the present invention also proposes a nuclear power plant fault diagnosis method.

[0061] As Figure 1 shown in, a nuclear power plant fault diagnosis method according to an embodiment of the present invention is shown, and the process mainly includes the following steps.

[0062] S101: Collect nuclear power plant data, and train and construct an edge-end model group and a cloud model group according to the nuclear power plant data, where the nuclear power plant data includes operation procedures, real-time operation data, and historical operation data.

[0063] Further, collect the historical operation data and real-time operation data of the sensors required in the nuclear power plant, train to obtain a principal component dimensionality reduction model, a latent variable regression model, and an independent feature extraction model, and integrate the principal component dimensionality reduction model, the latent variable regression model, and the independent feature extraction model and set a dynamic voter to construct an edge-end model group.

[0064] Further, a principal component dimensionality reduction model is constructed. As Figure 2 shown, first, for the input detected sensor data, perform mean standard deviation normalization to make the physical quantities collected by each sensor have similar weights. Denote the standardized sample data matrix as X, X ∈ R m*n , where m is the number of samples and n is the number of monitoring sensors. Convert the sample matrix X into its covariance matrix Σ, and perform eigenvalue decomposition. The expression of Σ is as follows:

[0065]

[0066] Σ = QDQ T ,

[0067] where the superscript T represents the transpose operation, Q is the eigenvector matrix, D is the eigenvalue diagonal matrix, and Q T is the transpose matrix of Q. According to the result of eigenvalue decomposition Σ, the k principal components are determined using the cumulative variance contribution rate method. Define the percentage of the information contained in the i-th principal component in the total information of the original data as PV i . If the first k principal components are selected to represent the states of all sensors, the expression for the percentage of the original information contained in the k principal components, that is, the cumulative variance contribution rate CPV of the principal components, is as follows:

[0068]

[0069] where λ i is the i-th eigenvalue of the covariance matrix, λ j is the j-th eigenvalue of the covariance matrix, k is the number of selected principal components, n is the total number of original variables, and PV is the variance contribution rate of the i-th principal component; when CPV exceeds the preset empirical value, it is judged that the principal components already contain sufficient information, and the selection of principal components is stopped. According to the selected principal components, the principal component loading matrix P and the residual space loading matrix P T and its principal component projection matrix C P and the residual space projection matrix C r are formed. The expression for the principal component projection matrix C P is as follows:

[0070] C P = P * P T ,

[0071] Calculate the statistic threshold TH1 according to the above formula and construct the principal component dimensionality reduction model. The threshold TH1 is composed of and Q a and is used as the subsequent fault criterion, where The expression is as follows:

[0072]

[0073] where k is the number of principal components, n is the number of training samples, α is the confidence level, and F α (k, n - k) represents the critical value of the F-test with the numerator and denominator degrees of freedom being k and n - k respectively at the significance level of α. At this time, the calculation expression for the threshold Q α is as follows:

[0074]

[0075] Among them, θ1 is the sum of the unextracted eigenvalue in the residual space, θ2 is the sum of the squares of the unextracted eigenvalues in the residual space, and c α is the critical value of the standard normal distribution at the significance level α, and h0 is the distribution approximation function in the residual space.

[0076] According to the real-time operation data and the residual space, calculate the T 2 statistic and Q statistic of the real-time operation data. The expression of the T 2 statistic is as follows:

[0077]

[0078] Among them, x is the sample vector, P is the principal component loading matrix, is the principal component eigenvalue matrix, and the expression of Q is as follows:

[0079]

[0080] Among them, x is the sample vector, C P is the principal component projection matrix, and C r is the residual space projection matrix.

[0081] In this embodiment, the standard threshold difference calculated by the principal component dimensionality reduction model is set to 0.3 * threshold TH1. If the T 2 or Q of a single sensor and the mean value of Q of the T 2 of other sensors are greater than or Q is greater than 3 * 0.3Q α , it is judged as a sensor failure. If the T 2 or Q of all sensors and the mean value of Q of the T 2 of the historical operation data are greater than or Q is greater than 3 * 0.3Q α , it is judged as a system failure.

[0082] Furthermore, construct a latent variable regression model, such as Figure 3 shown in to sort out and define the causal relationship of the real-time operation data of the sensors. The real-time operation data of the sensors that can be used as reasons is used as the dependent variable group, called Xc, and the real-time operation data of the sensors that are not convenient to be used as reasons is used as the result variable group, called Xe; Xc and Xe are standardized, and the standardization method is the same as that in the construction of the above principal component dimensionality reduction model, which will not be elaborated here.

[0083] For the fault diagnosis of sensor data, assume that there are q result variables that form the result variable group X e ={X e1 , X e2 , …, X eq}, and p dependent variables form the dependent variable group Xc = {X c1 , X c2 , …, X cp}, and its corresponding latent variables are denoted as U and T, which contain m and n latent elements respectively.

[0084] Initialize the residual matrix of the dependent variable group, denoted as Initialize the residual matrix of the outcome variable group, with the superscript std represents standardization, and construct the covariance matrices M c and M e respectively, where the expression of M c is as follows:

[0085]

[0086] M e The expression is as follows:

[0087]

[0088] Among them, M c reflects the comprehensive influence intensity of X c on X e , and M e reflects the reverse influence intensity of X e on X c .

[0089] Further solve the weight vectors of M c and M e , and the expressions are as follows:

[0090] w1 = eigvec(M c , λ max )(‖w1‖ = 1),

[0091] c1 = eigvec(M e , λ max )(‖c1‖ = 1),

[0092] Among them, w1 and c1 are the weight vectors of M c and M e respectively, eigvec is the function to calculate the eigenvector, λ max represents the largest eigenvalue of M c or M e , and ‖w1‖ = 1 and ‖c1‖ = 1 indicate that w1 and c1 are unit vectors.

[0093] According to w1 and c1, calculate the first group of latent variables t1 and u1, and their expressions are as follows:

[0094] t1 = E0w1,

[0095] u1 = F0c1,

[0096] Calculate X c The regression coefficient p1 of the latent variable t1, X e The regression coefficient q1 of the latent variable u1, and the regression weight β1 of u1 on t1, where the expression of the regression coefficient p1 is as follows:

[0097]

[0098] Among them, the denominator is a normalization factor, expressed as the square of the inner product of the latent variable t1; the numerator Calculates the projection of E0 in the direction of t1, and the overall expression calculates the standardized projection length of E0 in the direction of t1.

[0099] Furthermore, the expression of the regression coefficient q1 is as follows:

[0100]

[0101] Among them, the denominator is a normalization factor, expressed as the square of the inner product of the latent variable u1; the numerator Calculates the projection of F0 in the direction of u1, and the overall expression calculates the standardized projection length of F0 in the direction of u1.

[0102] The expression of the regression weight β1 of u1 on t1 is as follows:

[0103]

[0104] Among them, the numerator represents the covariance between two vectors, and the denominator is a normalization factor, expressed as the square of the inner product of the latent variable t1, and the overall expression calculates the expected change in u1 when t1 changes by one unit.

[0105] Update the residual matrix, and the expression is as follows:

[0106]

[0107] Among them, E1 and F1 respectively represent the updated X c The residual matrix E0 and the updated X e The residual matrix F0.

[0108] Repeat the above steps, and use cross-validation to determine the optimal number of components R, and the expression is as follows:

[0109]

[0110] Among them, K represents K-fold cross-validation, argmin rIndicates finding the value of r that minimizes the expression within the parentheses. Denotes the original dependent variable matrix of the test set in the k - fold cross - validation. Represents the predicted values of the k - fold test set by the model constructed with r components. Determine the threshold TH2 by and Q β which are used as subsequent fault criteria. The expressions therein are as follows:

[0111]

[0112] where R is the optimal number of components, m is the number of training samples, α is the confidence level, F α (R,m - R) represents the critical value of the F - test with degrees of freedom of the numerator and denominator being R and m - R respectively at the significance level α.

[0113] Furthermore, the calculation expression of Q β is as follows:

[0114]

[0115] where θ1 is the sum of the eigenvalues not extracted in the residual space, θ2 is the sum of the squares of the eigenvalues not extracted in the residual space, c α is the critical value of the standard normal distribution at the significance level α, and h0 is the distribution approximation function in the residual space.

[0116] The above steps are obtained from the training data, which are the data of the system in normal operation state. After training, the latent variables T and U are fixed, and the model is fixed. For the newly input operation data X test ={X testc , X teste}, calculate its statistics T 2 and Q.

[0117] According to the mean and standard deviation of Xc and Xe in the training stage, standardize X testc and X teste . Using the training matrix W = [w1, w2, w3…w R obtained in the training stage, calculate the latent variable score T test , and the calculation expression is as follows:

[0118] T test =X test W,

[0119] According to T test , calculate the T 2 statistic, and the calculation expression is as follows:

[0120]

[0121] where Λ is the covariance matrix of the latent variable scores of the training data, and the expression is:

[0122]

[0123] where m is the number of training samples.

[0124] Furthermore, calculate the Q statistic, and the calculation expression is as follows:

[0125]

[0126] where, is the standardized observed value of the i-th cause variable, is the predicted value of the i-th cause variable reconstructed by the latent variable model; is the standardized observed value of the j-th result variable, is the predicted value of the j-th result variable reconstructed by the latent variable model.

[0127] In this embodiment, the standard threshold difference calculated by the latent variable regression model is set to 0.4 * threshold TH2. If the T 2 or Q of a single sensor and the T 2 of other sensors or Q is greater than 3 * 0.4Q β , it is determined that the sensor is faulty. If the T 2 or Q of all sensors and the T 2 of the historical operation data or Q is greater than 3 * 0.4Q β , it is determined that the system is faulty.

[0128] Furthermore, construct an independent feature extraction model, as shown in Figure 4 . Take the real-time operation data of the sensor as the observation signal, denoted as X. Perform standardization processing on X, and the standardization method is the same as that in the construction of the above principal component reduction model, which will not be elaborated here, and denote it as Xnorm; whiten Xnorm, and the expression is as follows:

[0129]

[0130] where Z is the data matrix after whitening, the features are uncorrelated with each other and have equal variances, U is the eigenvector matrix, and U T is the transpose matrix of U, Λ is the diagonal matrix. Further, obtain the orthonormal matrix W. In this embodiment, a fixed-point iteration algorithm of negative entropy is adopted, and its expression is:

[0131]

[0132] where w new is the updated weight vector, m represents the number of samples, Z represents the above-mentioned whitening matrix, g represents the activation function, and a non-linear function is selected, with the expression:

[0133] g(u) = tanh(u),

[0134] w represents the weight vector, and g′ represents the derivative of the activation function.

[0135] Calculate the statistic threshold TH3 according to the above formula and construct an independent feature extraction model. The threshold TH3 consists of and Q γ and is used as the subsequent fault criterion. The calculation expression of

[0136]

[0137] is as follows: where R represents the number of independent features extracted, m is the number of samples, and F α (R, m - R) represents the critical value of the F-distribution with degrees of freedom (R, m - R) at the significance level α.

[0138] Q γ The calculation expression of is as follows:

[0139]

[0140] where θ1 is the sum of the eigenvalues not extracted in the residual space, θ2 is the sum of the squares of the eigenvalues not extracted in the residual space, c α is the critical value of the standard normal distribution at the significance level α, and h0 is the distribution approximation function in the residual space.

[0141] For the newly input operation data, after standardization, we get x new , and the expression after whitening is:

[0142]

[0143] where U test is the eigenvector matrix of the training data covariance matrix, and Λ test is the diagonal matrix of the eigenvalues of the training data covariance matrix. Based on this, the T 2 statistic can be calculated, and the calculation expression is as follows:

[0144]

[0145] where P train is the principal component loading matrix in the training stage. The calculation expression of the Q statistic is:

[0146]

[0147] Among them, x i is the i-th observation value, is the predicted value by the independent feature extraction model, and p is the total number of observed variables.

[0148] Furthermore, a dynamic voter is set. According to the threshold TH3 calculated in the above steps, T 2 and Q are calculated based on the real-time operation data to determine the data anomaly type. The data anomaly types include multi-sensor linear correlation anomaly, causal chain break, and non-Gaussian noise interference. If T 2 is greater than 2 times the threshold TH3 and Q is 1 times - 1.2 times Q in the threshold TH3 γ at this time, it is determined that the data anomaly type is multi-sensor linear correlation anomaly; if T 2 is 1 times - 1.2 times in the threshold TH3 γ and Q is greater than 3 times Q in the threshold TH3 2 at this time, it is determined that the data anomaly type is causal chain break; if T is 1 times - 1.4 times γ in the threshold TH3 and Q is 2 times - 3 times Q in the threshold TH3, it is determined that the data anomaly type is non-Gaussian noise interference.

[0149] Furthermore, the results of the constructed principal component dimensionality reduction model, latent variable regression model, and independent feature extraction model are normalized. In this embodiment, the min-max normalization method is used, and its expression is as follows:

[0150]

[0151] Among them, each model calculates its maximum value as max(x), minimum value as min(x), and Norm(x) is the result after normalization.

[0152] Furthermore, an operation procedure table is constructed through the operation procedures of the nuclear power plant, and a knowledge graph is constructed based on the operation procedure table. In this embodiment, the operation procedure table of the nuclear power plant is shown in Table 1 below.

[0153] Table 1 Operation Procedure Table of Nuclear Power Plant

[0154]

[0155]

[0156] Furthermore, according to the historical operation data, a deep learning confidence calculation model is trained. In this embodiment, a long short-term memory network (LSTM) is used as the deep learning confidence calculation model.

[0157] The deep learning confidence calculation model can be jointly represented by the following formula:

[0158]

[0159] Where f represents forgetting; i represents input; c represents memory, which is updated by forgetting; o represents output, which is used to generate output y and state h.

[0160] S102: Deploy the edge model group to the edge computing node near the equipment in the plant, and deploy the cloud model group to the cloud platform.

[0161] Specifically, install an edge segment gateway within a range of 2 to 4 meters around equipment such as centrifugal pump sets and compressor arrays. Deploy the edge model group to the edge segment gateway, and deploy the cloud model group in the server unit of the nuclear power plant. The edge network management accesses the intranet of the nuclear power plant so that the data processed by the edge model group can be transmitted to the cloud model group for decision-making.

[0162] S103: Input the real-time operation data of the nuclear power plant into the edge model group to detect the fault type. If the fault type is sensor abnormality, reconstruct the sensor parameters. If the fault type is system abnormality, upload the abnormality report to the cloud platform. The cloud model group performs reasoning and tracing on the abnormality report and pushes the fault information and decision-making plan.

[0163] Furthermore, by inputting the real-time operation data of the sensor into the edge model group, according to the type of real-time operation data, the dynamic voter selects the result of the optimal model as the detection result output. If the result obtained through the operation of the edge model group is greater than the threshold obtained during training, it is determined as abnormal and enters the fault type judgment stage.

[0164] Specifically, the mean standard deviation of the real-time operation data of the flow rate sensor is normalized by the edge model group, and then processed sequentially through three models: the principal component dimensionality reduction model, the latent variable regression model, and the independent feature extraction model, and the T of each model is obtained respectively 2 and Q. First, according to T 2 and Q and the and Q in the threshold TH a are compared. In this embodiment, through the calculation of the edge model group, the T of the flow rate sensor 2 is 2.2 times of then the dynamic voter directly determines it as multi-sensor linear correlation abnormality, selects the principal component dimensionality reduction model for anomaly detection, and the T of all sensors obtained 2 and the Q mean difference of the T of the historical operation data 2 is greater than It is determined as a system failure, and the real-time operation data and exception report at the time of failure are transmitted to the cloud platform. In other embodiments, if it is determined as a sensor failure, the average value is taken according to the historical operation data, the real-time operation data of the sensor is overwritten, and the data is re-transmitted to the edge model group.

[0165] Further, based on the real-time operation data and the exception report, the confidence level is calculated through the deep learning confidence level calculation model and weighted calculation to obtain the failure confidence level.

[0166] Further, the knowledge graph is used to reason the exception report to obtain a failure result set. In this embodiment, the failure result set obtained by reasoning according to the flow rate sensor is [rupture of the heat transfer tube of the steam generator, mis-opening of the feed water regulating valve] and the corresponding confidence levels [32%, 68%]. According to the failure result set, a solution set is inferred through the signed digraph model. In this embodiment, the solution set is [emergency shutdown, leakage location, closing the feed water regulating valve]; through the deep learning confidence level calculation model combined with the real-time operation data and the exception report, the corresponding confidence levels [18%, 22%, 60%] in the solution set are given, and finally the failure information pushed to the user is that the feed water regulating valve is mis-opened, and the decision-making plan is to close the feed water regulating valve.

[0167] Further, the failure is located on the relevant process system flow chart according to the failure information and pushed to the user.

[0168] It should be noted that for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations. However, those skilled in the art should know that this application is not limited by the described action sequence, because according to this application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0169] Based on another aspect of the embodiments of the present application, the present invention also provides a nuclear power plant fault diagnosis device. As Figure 5 shown, the device includes:

[0170] An acquisition and construction module 501, configured to acquire nuclear power plant data, train and construct an edge model group and a cloud model group according to the nuclear power plant data, where the nuclear power plant data includes operation procedures, real-time operation data, and historical operation data;

[0171] A module deployment module 502, configured to deploy the edge model group to an edge computing node near the equipment in the plant and deploy the cloud model group to the cloud platform;

[0172] The inference and decision-making module 503 is configured to input the real-time operation data of the nuclear power plant into the edge model group to detect the fault type. If the fault type is sensor anomaly, the sensor parameters are reconstructed. If the fault type is system anomaly, the anomaly report is uploaded to the cloud platform, and the cloud model group is used to infer and trace the anomaly report, and push the fault information and decision-making scheme.

[0173] As an alternative solution, the above device is further configured to: train based on the real-time operation data and the historical operation data to obtain a principal component dimensionality reduction model, a latent variable regression model, and an independent feature extraction model, and obtain the edge model group through model integration.

[0174] As an alternative solution, the above device is further configured to: obtain the knowledge graph of the nuclear power plant according to the operation regulations, train based on the real-time operation data and the historical operation data to obtain a signed directed graph model and a deep learning confidence calculation model, and obtain the cloud model group through model integration.

[0175] As an alternative solution, the above device is further configured to: set a dynamic voter in the edge model group, and the dynamic voter makes a decision based on the results of the principal component dimensionality reduction model, the latent variable regression model, and the independent feature extraction model.

[0176] As an alternative solution, the above device is further configured to: after inputting the real-time operation data of the nuclear power plant, normalize the results of the principal component dimensionality reduction model, the latent variable regression model, and the independent feature extraction model.

[0177] As an alternative solution, the above device is further configured to: input the historical operation data into the edge model group to obtain the threshold TH of each model respectively, and set 0.25TH - 0.6TH as the standard threshold difference.

[0178] As an alternative solution, the above device is further configured to: judge for the real-time operation data. If the absolute value of the difference between the threshold TH of a single sensor and the mean value of the threshold TH of other sensors exceeds 3 times the standard threshold difference, it is determined as a sensor anomaly; if the absolute value of the difference between the threshold TH of all sensors and the threshold TH calculated based on the historical normal operation data exceeds 3 times the standard threshold difference, it is determined as a system anomaly; when the output results of most models determine the same anomaly type, it is directly determined as that anomaly type, and when the output results of each model are different, the anomaly type determined by the model with the highest weight is selected.

[0179] As an alternative solution, the above device is further configured to: correct the sensor parameters based on the historical operation data.

[0180] As an alternative, the above device is further configured to: re - input the corrected sensor parameters into the edge - side model group.

[0181] As an alternative, the above device is further configured to: calculate the confidence through the deep - learning confidence calculation model based on the real - time operation data and the exception report; infer a fault result set through the knowledge graph for the exception report, and based on the fault result set, infer a solution set through the signed - digraph model; determine a fault from the fault result set through the deep - learning confidence calculation model.

[0182] As an alternative, the above device is further configured to: calculate and weight the faults in the fault result set through the deep - learning confidence calculation model to obtain the confidence.

[0183] As an alternative, the above device is further configured to: push the fault information with the highest confidence in the fault result set and its corresponding decision scheme in the solution set to the front - end, and perform fault location on the relevant process system flow chart.

[0184] In the embodiments of the present application, the term "module" or "unit" refers to a computer program with a predetermined function or a part of a computer program, which works together with other related parts to achieve a predetermined goal, and can be fully or partially implemented by using software, hardware (such as a processing circuit or a memory), or a combination thereof. Similarly, one processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be a part of an overall module or unit that includes the function of the module or unit.

[0185] Regarding the device in the above embodiments, the specific manner in which each module performs operations has been described in detail in the embodiments related to the method, and will not be elaborated here in detail.

[0186] According to one aspect of the present application, a computer program product is provided, and the computer program product includes a computer program.

[0187] The serial numbers of the above embodiments of the present application are only for description and do not represent the advantages or disadvantages of the embodiments.

[0188] Figure 6 A computer system block diagram of an electronic device for implementing the embodiments of the present application is schematically shown.

[0189] It should be noted that Figure 6The computer system 600 of the illustrated electronic device is merely an example and should not impose any limitation on the functions and scope of use of the embodiments of the present application.

[0190] As Figure 6 shown, the computer system 600 includes a central processing unit 601 (CPU), which can perform various appropriate actions and processes according to the program stored in the read-only memory 602 (ROM) or the program loaded from the storage section 608 into the random access memory 603 (RAM). In the random access memory 603, various programs and data required for system operation are also stored. The central processing unit 601, the read-only memory 602, and the random access memory 603 are connected to each other via a bus 604. The input / output interface 605 (Input / Output interface, i.e., I / O interface) is also connected to the bus 604.

[0191] The following components are connected to the input / output interface 605: an input section 606 including a keyboard, a mouse, etc.; an output section 607 including, for example, a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a local area network card, a modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the input / output interface 605 as needed. A removable medium 63, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 610 as needed so that a computer program read from it can be installed into the storage section 608 as needed.

[0192] In particular, according to the embodiments of the present application, the processes described in each method flowchart can be implemented as computer software programs. For example, the embodiments of the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes program codes for performing the methods shown in the flowcharts. In such an embodiment, the computer program can be downloaded and installed from the network via the communication section 609, and / or installed from the removable medium 63. When the computer program is executed by the central processing unit 601, various functions defined in the system of the present application are executed.

[0193] In such an embodiment, the computer program can be downloaded and installed from a network through the communication section 609, and / or installed from the removable medium 63. When the computer program is executed by the central processing unit 601, various functions provided by the embodiments of the present application are executed.

[0194] According to another aspect of the embodiments of the present application, an electronic device for a nuclear power plant fault diagnosis method is further provided. In this embodiment, the electronic device is taken as an example of a terminal device for illustration. As Figure 7 shown, the electronic device includes a memory 702 and a processor 704. A computer program is stored in the memory 702, and the processor 704 is configured to execute the steps in any of the above method embodiments through the computer program.

[0195] Optionally, in this embodiment, the above-mentioned electronic device may be at least one network device among multiple network devices of a computer network.

[0196] Optionally, in this embodiment, the above-mentioned processor may be configured to execute the methods in the embodiments of the present application through a computer program.

[0197] Optionally, those of ordinary skill in the art can understand that Figure 7 the structure shown is only schematic, Figure 7 and it does not limit the structure of the above-mentioned electronic device. For example, the electronic device may further include more or fewer components (such as a network interface, etc.) than those shown Figure 7 in the figure, or have a different configuration from that shown Figure 7 in the figure.

[0198] Among them, the memory 702 can be used to store software programs and modules, such as program instructions / modules corresponding to a nuclear power plant fault diagnosis method and device in the embodiments of the present application. The processor 704 executes various functional applications and data processing by running the software programs and modules stored in the memory 702, that is, implements the above-mentioned nuclear power plant fault diagnosis method. The memory 702 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some instances, the memory 702 may further include a memory remotely disposed relative to the processor 704, and these remote memories can be connected to the terminal through a network. Examples of the above-mentioned network include but are not limited to the Internet, enterprise intranets, local area networks, mobile communication networks, and their combinations. Among them, the memory 702 can specifically but not limitedly be used to store the collected real-time operation data or cleaned data information. As an example, as Figure 7As shown, the above-mentioned memory 702 may, but is not limited to, include the acquisition and construction module 501, the module deployment module 502, and the inference and decision-making module 503 in the above-mentioned nuclear power plant fault diagnosis device. In addition, it may also include, but is not limited to, other module units in the above-mentioned device, which will not be elaborated in this example.

[0199] Optionally, the above-mentioned transmission device 706 is used to receive or send data via a network. Specific examples of the above-mentioned network may include a wired network and a wireless network. In one example, the transmission device 706 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices and routers through a network cable so as to communicate with the Internet or a local area network. In one example, the transmission device 706 is a Radio Frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0200] In addition, the above-mentioned electronic device further includes: a display 708, which is used to display the above-mentioned real-time operation data or cleaning data; and a connection bus 710, which is used to connect each module component in the above-mentioned electronic device.

[0201] In other embodiments, the above-mentioned terminal device or server may be a node in a distributed system. Among them, the distributed system may be a blockchain system, and the blockchain system may be a distributed system formed by connecting the multiple nodes in the form of network communication. Among them, the nodes can form a peer-to-peer network, and any form of computing device, such as an electronic device such as a server or a terminal, can become a node in the blockchain system by joining the peer-to-peer network.

[0202] According to one aspect of the present application, there is provided a computer-readable storage medium. The processor of the electronic device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the electronic device executes a nuclear power plant fault diagnosis method provided in various optional implementation manners of the above-mentioned nuclear power plant fault diagnosis method.

[0203] Optionally, in this embodiment, the above-mentioned computer-readable storage medium may be set to store the programs for executing the methods in the various embodiments of the present application.

[0204] Optionally, in this embodiment, those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the relevant hardware of the terminal device through a program, and this program can be stored in a computer-readable storage medium. The storage medium may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, an optical disk, or the like.

[0205] The serial numbers of the embodiments of the present application above are only for description and do not represent the superiority or inferiority of the embodiments.

[0206] If the integrated unit in the above embodiments is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in the above computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing one or more electronic devices to execute all or part of the steps of the methods described in the various embodiments of the present application.

[0207] In the above embodiments of the present application, the descriptions of the various embodiments have their own emphases. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0208] In the several embodiments provided by the present application, it should be understood that the disclosed application program can be implemented in other ways. Among them, the device embodiments described above are only illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some interfaces. The indirect coupling or communication connection of units or modules can be in an electrical or other form.

[0209] The unit described as a separated component may or may not be physically separated. The component displayed as a unit may or may not be a physical unit, that is, it may be located in one place, or may be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0210] In addition, in each embodiment of the present application, each functional unit can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0211] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

[0212] In summary, as can be seen from the above description, the above embodiments of the present invention achieve the following technical effects:

[0213] 1. The present invention provides a nuclear power plant fault diagnosis method and device. Through the collaborative architecture of the cloud and the edge, the advantages of the edge and the cloud are fully utilized, achieving the high efficiency and accuracy of nuclear power plant fault diagnosis; the edge model group is responsible for real-time data collection and preliminary processing, which can quickly respond to abnormal situations, reduce data transmission delay, and ensure the real-time nature of fault monitoring; the cloud model group, relying on its powerful computing and storage capabilities, conducts in-depth analysis and model training on complex faults to provide high-precision diagnostic results; the present invention also integrates an expert knowledge reasoning model and a variety of data-driven algorithms, which can analyze and verify faults from different angles, quickly locate possible fault types and fault sources; the data-driven model can effectively monitor abnormal data, distinguish sensor faults from system faults, and further evaluate the confidence level of the fault source, ultimately realizing the accurate traceability of faults; this method of multi-model fusion not only improves the accuracy of fault diagnosis, but also can analyze the fault propagation path, predict the fault evolution trend, and provide a scientific basis for taking preventive measures in advance.

[0214] 2. The present invention provides a nuclear power plant fault diagnosis method and device. By optimizing the data processing process and resource allocation through a variety of data-driven algorithm models (such as PCA, PLS, ICA), the reliability and scalability of the system are significantly improved; on the one hand, after preprocessing the collected data, the edge model group only uploads key data to the cloud, effectively reducing the data transmission volume, reducing the network bandwidth pressure and data transmission delay, thereby improving the overall reliability of the system; on the other hand, the collaborative architecture of the cloud model group enables the system to flexibly adjust resource allocation according to the scale and data volume of the nuclear power plant, further calculating the fault confidence level through deep learning models (such as LSTM). The edge sensor devices can be flexibly arranged according to actual needs, while the cloud resources can be dynamically expanded according to computing and storage requirements, adapting to the application scenarios of nuclear power plants of different scales, showing good scalability.

[0215] 3. The present invention provides a nuclear power plant fault diagnosis method and device. By constructing an auxiliary decision-making knowledge graph and combining the operation data and operation procedures of the nuclear power plant, corresponding treatment strategies and procedures can be quickly generated based on the fault diagnosis results, providing intuitive and scientific decision-making references for operators. The system uses a cloud model group for in-depth analysis and pushes decision-making information to the main control room and on-site maintenance personnel in real time to ensure information synchronization and improve decision-making efficiency and response speed. In addition, the system also has functions of fault traceability and fault propagation path analysis, which can help operators anticipate the fault evolution trend in advance and thus take more reasonable countermeasures. The combination of such real-time monitoring, early warning and auxiliary decision-making functions can significantly reduce the probability of unplanned reactor shutdown in the nuclear power plant, reduce economic losses, and ensure the safe and stable operation of the nuclear power plant.

[0216] 4. The present invention provides a nuclear power plant fault diagnosis method and device. By clearly defining the process division of labor between the edge model group and the cloud model group, the problem of resource waste in the traditional centralized or single-edge computing mode is avoided. The edge model group focuses on real-time data processing, and the cloud model group is responsible for complex analysis and model training. This division of labor makes resource utilization more efficient. At the same time, the system supports flexibly selecting to deploy the data-driven model of the edge model group to the edge or the cloud according to the needs of the monitoring object, further improving the flexibility and adaptability of the system. This flexible deployment strategy can not only meet the needs of different application scenarios but also dynamically adjust resource allocation according to the actual operation situation to ensure that the system can operate efficiently under different working conditions.

[0217] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

[0218] It should be noted that in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variation thereof is intended to cover a non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising one..." does not exclude the presence of another identical element in the process, method, article or device comprising the element.

[0219] It should be noted that in the description of this specification, the descriptions with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples", etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

Claims

1. A nuclear power plant fault diagnosis method, characterized in that, Including: Collecting nuclear power plant data, training and constructing an edge - side model group and a cloud - side model group according to the nuclear power plant data, where the nuclear power plant data includes operation procedures, real - time operation data, and historical operation data; Deploying the edge - side model group to edge computing nodes near the equipment in the plant building, and deploying the cloud - side model group to the cloud platform; Inputting the real - time operation data of the nuclear power plant into the edge - side model group to detect the fault type. If the fault type is sensor anomaly, reconstruct the sensor parameters. If the fault type is system anomaly, upload the anomaly report to the cloud platform, and the cloud - side model group performs reasoning and tracing on the anomaly report, and pushes the fault information and decision - making solutions.

2. The method according to claim 1, wherein Training and constructing the edge - side model group and the cloud - side model group according to the nuclear power plant data includes: Training based on the real - time operation data and the historical operation data to obtain a principal component dimensionality reduction model, a latent variable regression model, and an independent feature extraction model, and obtaining the edge - side model group through model integration.

3. The method according to claim 1, wherein Training and constructing the edge - side model group and the cloud - side model group according to the nuclear power plant data further includes: Obtaining a nuclear power plant knowledge graph according to the operation procedures, training based on the real - time operation data and the historical operation data to obtain a signed digraph model and a deep - learning confidence calculation model, and obtaining the cloud - side model group through model integration.

4. The method according to claim 2, wherein And obtaining the edge - side model group through model integration includes: Setting a dynamic voter in the edge - side model group, and the dynamic voter makes decisions based on the results of the principal component dimensionality reduction model, the latent variable regression model, and the independent feature extraction model.

5. The method according to claim 4, wherein The dynamic voter makes decisions based on the results of the principal component dimensionality reduction model, the latent variable regression model, and the independent feature extraction model, including: After inputting the real - time operation data of the nuclear power plant, normalizing the results of the principal component dimensionality reduction model, the latent variable regression model, and the independent feature extraction model.

6. The method according to claim 4, characterized in that The dynamic voter makes decisions based on the results of the principal component dimensionality reduction model, the latent variable regression model, and the independent feature extraction model further includes: Inputting the historical operation data into the edge - side model group to obtain the threshold TH of each model respectively, and setting 0.25TH - 0.6TH as the standard threshold difference.

7. The method according to claim 6, wherein The dynamic voter makes decisions based on the results of the principal component dimensionality reduction model, the latent variable regression model, and the independent feature extraction model, including: Judging for the real - time operation data. If the absolute value of the difference between the threshold TH of a single sensor and the mean value of the threshold TH of other sensors exceeds 3 times the standard threshold difference, it is determined as a sensor anomaly; if the absolute value of the difference between the threshold TH of all sensors and the threshold TH calculated based on historical normal operation data exceeds 3 times the standard threshold difference, it is determined as a system anomaly; When the output results of most models determine the same anomaly type, it is directly determined as that anomaly type. When the output results of each model are all different, select the anomaly type determined by the model with the highest weight.

8. The method according to claim 1, wherein If the fault type is sensor anomaly, reconstruct the sensor parameters, including: Based on the historical operation data, correct the sensor parameters.

9. The method according to claim 8, wherein After correcting the sensor parameters, further include: Re-input the corrected sensor parameters into the edge model group.

10. The method according to claim 3, characterized in that, Infer and trace the anomaly report through the cloud model group, including: Based on the real-time operation data and the anomaly report, calculate the confidence level through the deep learning confidence calculation model; Infer the fault result set through the knowledge graph for the anomaly report, and based on the fault result set, infer the solution set through the signed directed graph model; Determine the fault from the fault result set through the deep learning confidence calculation model.

11. The method according to claim 10, wherein Determine the fault from the fault result set through the deep learning confidence calculation model, including: Through the deep learning confidence calculation model, calculate and weight the faults in the fault result set to obtain the confidence level.

12. The method according to claim 11, wherein Push the fault information and decision-making solutions, including: Push the fault information with the highest confidence level in the fault result set and its corresponding decision-making solution in the solution set to the front end, and perform fault location on the relevant process system flow chart.

13. A nuclear power plant fault diagnosis device, characterized in that, Include: An acquisition and construction module, configured to acquire nuclear power plant data, train and construct an edge model group and a cloud model group according to the nuclear power plant data, where the nuclear power plant data includes operation procedures, real-time operation data, and historical operation data; A module deployment module, configured to deploy the edge model group to an edge computing node near the equipment in the plant building, and deploy the cloud model group to a cloud platform; An inference and decision-making module, configured to input the real-time operation data of the nuclear power plant into the edge model group to detect the fault type. If the fault type is sensor anomaly, reconstruct the sensor parameters. If the fault type is system anomaly, upload the anomaly report to the cloud platform, infer and trace the anomaly report through the cloud model group, and push the fault information and decision-making solutions.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein the computer program can execute the method described in any one of claims 1 to 12 when being run by an electronic device.

15. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method described in any one of claims 1 to 12.

16. An electronic device, comprising a memory and a processor, characterized in that, A computer program is stored in the memory, and the processor is configured to execute the method described in any one of claims 1 to 12 through the computer program.

Images